5fb84a02f0
- Add standalone training binaries: TGGN, KAN, xLSTM, Diffusion (DBN data) - Update Dockerfile.training: 6 → 16 binaries (all 10 models + hyperopt + baseline) - Expand train.sh: 4 → 10 models, fix registry URL and GPU pool nodeSelector - Add GPU overlay manifests for trading-service and ml-training-service - Create training data PVC and upload pod manifests - Expand web-gateway model validation: 4 → 10 types (training + tune routes) - Extend dashboard: 10 model cards grouped by category (RL/Temporal/Graph/Generative) - Add training image build job to Gitea CI workflow - Update GPU taint controller to exclude inference pool from tainting - Fix job-template nodeSelector: gpu → gpu-training Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
93 lines
2.5 KiB
HCL
93 lines
2.5 KiB
HCL
resource "scaleway_vpc_private_network" "foxhunt" {
|
|
name = "${var.cluster_name}-pn"
|
|
region = var.region
|
|
}
|
|
|
|
resource "scaleway_k8s_cluster" "foxhunt" {
|
|
name = var.cluster_name
|
|
version = var.k8s_version
|
|
cni = "cilium"
|
|
region = var.region
|
|
delete_additional_resources = true
|
|
private_network_id = scaleway_vpc_private_network.foxhunt.id
|
|
|
|
auto_upgrade {
|
|
enable = true
|
|
maintenance_window_start_hour = 4
|
|
maintenance_window_day = "sunday"
|
|
}
|
|
|
|
autoscaler_config {
|
|
disable_scale_down = false
|
|
scale_down_delay_after_add = "10m"
|
|
scale_down_unneeded_time = "10m"
|
|
estimator = "binpacking"
|
|
ignore_daemonsets_utilization = true
|
|
}
|
|
}
|
|
|
|
resource "scaleway_k8s_pool" "always_on" {
|
|
cluster_id = scaleway_k8s_cluster.foxhunt.id
|
|
name = "always-on"
|
|
node_type = var.always_on_type
|
|
size = 1
|
|
min_size = 1
|
|
max_size = 1
|
|
autoscaling = false
|
|
autohealing = true
|
|
region = var.region
|
|
}
|
|
|
|
resource "scaleway_k8s_pool" "ci" {
|
|
cluster_id = scaleway_k8s_cluster.foxhunt.id
|
|
name = "ci"
|
|
node_type = var.ci_type
|
|
size = 1
|
|
min_size = 0
|
|
max_size = var.ci_max_size
|
|
autoscaling = true
|
|
autohealing = true
|
|
region = var.region
|
|
|
|
lifecycle {
|
|
ignore_changes = [size]
|
|
}
|
|
}
|
|
|
|
# GPU pool for ML training (H100 — large VRAM for ensemble training)
|
|
resource "scaleway_k8s_pool" "gpu_training" {
|
|
count = var.enable_gpu_training_pool ? 1 : 0
|
|
cluster_id = scaleway_k8s_cluster.foxhunt.id
|
|
name = "gpu-training"
|
|
node_type = var.gpu_training_type
|
|
size = 1
|
|
min_size = 0
|
|
max_size = var.gpu_training_max_size
|
|
autoscaling = true
|
|
autohealing = true
|
|
region = var.region
|
|
|
|
lifecycle {
|
|
ignore_changes = [size]
|
|
}
|
|
}
|
|
|
|
# GPU pool for inference during trading (L4 — cost-effective for forward passes)
|
|
# When trading: trading_service + ml_training_service move here via GPU-enabled manifests
|
|
resource "scaleway_k8s_pool" "gpu_inference" {
|
|
count = var.enable_gpu_inference_pool ? 1 : 0
|
|
cluster_id = scaleway_k8s_cluster.foxhunt.id
|
|
name = "gpu-inference"
|
|
node_type = var.gpu_inference_type
|
|
size = 1
|
|
min_size = 0
|
|
max_size = var.gpu_inference_max_size
|
|
autoscaling = true
|
|
autohealing = true
|
|
region = var.region
|
|
|
|
lifecycle {
|
|
ignore_changes = [size]
|
|
}
|
|
}
|